Information Technology Reference
In-Depth Information
=
The
variational
formulation
( 12.10 )
then
reads:
For m
1 ,...,M ,
find
(U m,j ) r j = 0
V r m + 1
L
such that for all (W m,i ) r m
V r m + 1
L
0
the following holds:
i
=
r m
r m
r m
k m
2
C ij (U m,j ,W m,i )
I ij a(U m,j ,W m,i )
+
=
f m,i ,
i,j
=
0
i,j
=
0
i
=
0
ϕ i
where f m,i =
(
1 )(U m 1 (t m 1 ), W m,i ) , with U 0 (t 0 )
=
u L, 0
V L , and for i, j
=
1 ,...,r m ,
1
1
1
1
C ij =
ϕ j
ϕ j
ϕ i
I ij =
t
t
ϕ j d
+
(
1 )
(
1 ),
ϕ j
ϕ i d
=
δ ij .
The matrices C m and I m , m
1 ,...,M , are independent of the time step and can be
calculated in a preprocessing step. Their size, however, depends on the correspond-
ing approximation order r m .
Denoting by M and A the mass and (wavelet compressed) stiffness matrix with
respect to (
=
·
,
·
) and a(
·
,
·
) ,( 12.10 ) takes the matrix form
(r m + 1 )N L
Find u m
∈ R
=
such that for m
1 ,...,M
C m
A u m
k
2 I m
( ϕ m
M ) u m 1 ,
(12.13)
M
+
=
u 0
=
u 0 ,
ϕ 1
where u m
and ϕ m
denotes the coefficient vector of U
| J m
:=
(
(
1 ),...,
ϕ r m + 1 (
1 )) ∈ R
r m + 1 . Furthermore, u 0 ∈ R
N L
is the coefficient vector of u L, 0
with respect to the wavelet basis of V L .
For notational simplicity, we consider for the rest of this section a generic time
step, omit the index m and write C and I for the matrices C m and I m , respectively,
u , ϕ for the vectors appearing in ( 12.13 ), and r for the approximation order. Fur-
thermore, we denote the right hand side by f = ( ϕ m
M ) u m 1 .
12.3.2 Solution Algorithm
The system ( 12.13 )ofsize (r
+
1 )N L can be reduced to solving r
+
1 linear systems
QTQ be the Schur decomposition of the (r
of size N L . To this end, let C
=
+
1 )
×
(r +
1 ) matrix C with a unitary matrix Q and an upper triangular matrix T . Note that
the diagonal of T contains the eigenvalues λ 1 2 ,...,λ r + 1 of C . Then, multiplying
( 12.13 )by Q
I from the left results in the linear system
T
A W
k
2 I
( Q
( Q
M
+
=
g
with
w
=
I ) u , g
=
I ) f .
Search WWH ::




Custom Search